5b371109ec1cdf5fe80cf24e4e53f9ee60018af8
[linux-2.6.git] / net / sched / sch_prio.c
1 /*
2  * net/sched/sch_prio.c Simple 3-band priority "scheduler".
3  *
4  *              This program is free software; you can redistribute it and/or
5  *              modify it under the terms of the GNU General Public License
6  *              as published by the Free Software Foundation; either version
7  *              2 of the License, or (at your option) any later version.
8  *
9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11  *              Init --  EINVAL when opt undefined
12  */
13
14 #include <linux/module.h>
15 #include <asm/uaccess.h>
16 #include <asm/system.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/mm.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/in.h>
25 #include <linux/errno.h>
26 #include <linux/interrupt.h>
27 #include <linux/if_ether.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/notifier.h>
32 #include <net/ip.h>
33 #include <net/route.h>
34 #include <linux/skbuff.h>
35 #include <net/sock.h>
36 #include <net/pkt_sched.h>
37
38
39 struct prio_sched_data
40 {
41         int bands;
42         struct tcf_proto *filter_list;
43         u8  prio2band[TC_PRIO_MAX+1];
44         struct Qdisc *queues[TCQ_PRIO_BANDS];
45 };
46
47
48 static struct Qdisc *
49 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
50 {
51         struct prio_sched_data *q = qdisc_priv(sch);
52         u32 band = skb->priority;
53         struct tcf_result res;
54
55         *qerr = NET_XMIT_BYPASS;
56         if (TC_H_MAJ(skb->priority) != sch->handle) {
57 #ifdef CONFIG_NET_CLS_ACT
58                 switch (tc_classify(skb, q->filter_list, &res)) {
59                 case TC_ACT_STOLEN:
60                 case TC_ACT_QUEUED:
61                         *qerr = NET_XMIT_SUCCESS;
62                 case TC_ACT_SHOT:
63                         return NULL;
64                 };
65
66                 if (!q->filter_list ) {
67 #else
68                 if (!q->filter_list || tc_classify(skb, q->filter_list, &res)) {
69 #endif
70                         if (TC_H_MAJ(band))
71                                 band = 0;
72                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
73                 }
74                 band = res.classid;
75         }
76         band = TC_H_MIN(band) - 1;
77         if (band > q->bands)
78                 return q->queues[q->prio2band[0]];
79
80         return q->queues[band];
81 }
82
83 static int
84 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
85 {
86         struct Qdisc *qdisc;
87         int ret;
88
89         qdisc = prio_classify(skb, sch, &ret);
90 #ifdef CONFIG_NET_CLS_ACT
91         if (qdisc == NULL) {
92
93                 if (ret == NET_XMIT_BYPASS)
94                         sch->qstats.drops++;
95                 kfree_skb(skb);
96                 return ret;
97         }
98 #endif
99
100         if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
101                 sch->bstats.bytes += skb->len;
102                 sch->bstats.packets++;
103                 sch->q.qlen++;
104                 return NET_XMIT_SUCCESS;
105         }
106         sch->qstats.drops++;
107         return ret;
108 }
109
110
111 static int
112 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
113 {
114         struct Qdisc *qdisc;
115         int ret;
116
117         qdisc = prio_classify(skb, sch, &ret);
118 #ifdef CONFIG_NET_CLS_ACT
119         if (qdisc == NULL) {
120                 if (ret == NET_XMIT_BYPASS)
121                         sch->qstats.drops++;
122                 kfree_skb(skb);
123                 return ret;
124         }
125 #endif
126
127         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
128                 sch->q.qlen++;
129                 sch->qstats.requeues++;
130                 return 0;
131         }
132         sch->qstats.drops++;
133         return NET_XMIT_DROP;
134 }
135
136
137 static struct sk_buff *
138 prio_dequeue(struct Qdisc* sch)
139 {
140         struct sk_buff *skb;
141         struct prio_sched_data *q = qdisc_priv(sch);
142         int prio;
143         struct Qdisc *qdisc;
144
145         for (prio = 0; prio < q->bands; prio++) {
146                 qdisc = q->queues[prio];
147                 skb = qdisc->dequeue(qdisc);
148                 if (skb) {
149                         sch->q.qlen--;
150                         return skb;
151                 }
152         }
153         return NULL;
154
155 }
156
157 static unsigned int prio_drop(struct Qdisc* sch)
158 {
159         struct prio_sched_data *q = qdisc_priv(sch);
160         int prio;
161         unsigned int len;
162         struct Qdisc *qdisc;
163
164         for (prio = q->bands-1; prio >= 0; prio--) {
165                 qdisc = q->queues[prio];
166                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
167                         sch->q.qlen--;
168                         return len;
169                 }
170         }
171         return 0;
172 }
173
174
175 static void
176 prio_reset(struct Qdisc* sch)
177 {
178         int prio;
179         struct prio_sched_data *q = qdisc_priv(sch);
180
181         for (prio=0; prio<q->bands; prio++)
182                 qdisc_reset(q->queues[prio]);
183         sch->q.qlen = 0;
184 }
185
186 static void
187 prio_destroy(struct Qdisc* sch)
188 {
189         int prio;
190         struct prio_sched_data *q = qdisc_priv(sch);
191         struct tcf_proto *tp;
192
193         while ((tp = q->filter_list) != NULL) {
194                 q->filter_list = tp->next;
195                 tcf_destroy(tp);
196         }
197
198         for (prio=0; prio<q->bands; prio++)
199                 qdisc_destroy(q->queues[prio]);
200 }
201
202 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
203 {
204         struct prio_sched_data *q = qdisc_priv(sch);
205         struct tc_prio_qopt *qopt = RTA_DATA(opt);
206         int i;
207
208         if (opt->rta_len < RTA_LENGTH(sizeof(*qopt)))
209                 return -EINVAL;
210         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
211                 return -EINVAL;
212
213         for (i=0; i<=TC_PRIO_MAX; i++) {
214                 if (qopt->priomap[i] >= qopt->bands)
215                         return -EINVAL;
216         }
217
218         sch_tree_lock(sch);
219         q->bands = qopt->bands;
220         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
221
222         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
223                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
224                 if (child != &noop_qdisc) {
225                         qdisc_tree_decrease_qlen(child, child->q.qlen);
226                         qdisc_destroy(child);
227                 }
228         }
229         sch_tree_unlock(sch);
230
231         for (i=0; i<q->bands; i++) {
232                 if (q->queues[i] == &noop_qdisc) {
233                         struct Qdisc *child;
234                         child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
235                                                   TC_H_MAKE(sch->handle, i + 1));
236                         if (child) {
237                                 sch_tree_lock(sch);
238                                 child = xchg(&q->queues[i], child);
239
240                                 if (child != &noop_qdisc) {
241                                         qdisc_tree_decrease_qlen(child,
242                                                                  child->q.qlen);
243                                         qdisc_destroy(child);
244                                 }
245                                 sch_tree_unlock(sch);
246                         }
247                 }
248         }
249         return 0;
250 }
251
252 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
253 {
254         struct prio_sched_data *q = qdisc_priv(sch);
255         int i;
256
257         for (i=0; i<TCQ_PRIO_BANDS; i++)
258                 q->queues[i] = &noop_qdisc;
259
260         if (opt == NULL) {
261                 return -EINVAL;
262         } else {
263                 int err;
264
265                 if ((err= prio_tune(sch, opt)) != 0)
266                         return err;
267         }
268         return 0;
269 }
270
271 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
272 {
273         struct prio_sched_data *q = qdisc_priv(sch);
274         unsigned char *b = skb_tail_pointer(skb);
275         struct tc_prio_qopt opt;
276
277         opt.bands = q->bands;
278         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
279         RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
280         return skb->len;
281
282 rtattr_failure:
283         skb_trim(skb, b - skb->data);
284         return -1;
285 }
286
287 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
288                       struct Qdisc **old)
289 {
290         struct prio_sched_data *q = qdisc_priv(sch);
291         unsigned long band = arg - 1;
292
293         if (band >= q->bands)
294                 return -EINVAL;
295
296         if (new == NULL)
297                 new = &noop_qdisc;
298
299         sch_tree_lock(sch);
300         *old = q->queues[band];
301         q->queues[band] = new;
302         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
303         qdisc_reset(*old);
304         sch_tree_unlock(sch);
305
306         return 0;
307 }
308
309 static struct Qdisc *
310 prio_leaf(struct Qdisc *sch, unsigned long arg)
311 {
312         struct prio_sched_data *q = qdisc_priv(sch);
313         unsigned long band = arg - 1;
314
315         if (band >= q->bands)
316                 return NULL;
317
318         return q->queues[band];
319 }
320
321 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
322 {
323         struct prio_sched_data *q = qdisc_priv(sch);
324         unsigned long band = TC_H_MIN(classid);
325
326         if (band - 1 >= q->bands)
327                 return 0;
328         return band;
329 }
330
331 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
332 {
333         return prio_get(sch, classid);
334 }
335
336
337 static void prio_put(struct Qdisc *q, unsigned long cl)
338 {
339         return;
340 }
341
342 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
343 {
344         unsigned long cl = *arg;
345         struct prio_sched_data *q = qdisc_priv(sch);
346
347         if (cl - 1 > q->bands)
348                 return -ENOENT;
349         return 0;
350 }
351
352 static int prio_delete(struct Qdisc *sch, unsigned long cl)
353 {
354         struct prio_sched_data *q = qdisc_priv(sch);
355         if (cl - 1 > q->bands)
356                 return -ENOENT;
357         return 0;
358 }
359
360
361 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
362                            struct tcmsg *tcm)
363 {
364         struct prio_sched_data *q = qdisc_priv(sch);
365
366         if (cl - 1 > q->bands)
367                 return -ENOENT;
368         tcm->tcm_handle |= TC_H_MIN(cl);
369         if (q->queues[cl-1])
370                 tcm->tcm_info = q->queues[cl-1]->handle;
371         return 0;
372 }
373
374 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
375                                  struct gnet_dump *d)
376 {
377         struct prio_sched_data *q = qdisc_priv(sch);
378         struct Qdisc *cl_q;
379
380         cl_q = q->queues[cl - 1];
381         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
382             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
383                 return -1;
384
385         return 0;
386 }
387
388 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
389 {
390         struct prio_sched_data *q = qdisc_priv(sch);
391         int prio;
392
393         if (arg->stop)
394                 return;
395
396         for (prio = 0; prio < q->bands; prio++) {
397                 if (arg->count < arg->skip) {
398                         arg->count++;
399                         continue;
400                 }
401                 if (arg->fn(sch, prio+1, arg) < 0) {
402                         arg->stop = 1;
403                         break;
404                 }
405                 arg->count++;
406         }
407 }
408
409 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
410 {
411         struct prio_sched_data *q = qdisc_priv(sch);
412
413         if (cl)
414                 return NULL;
415         return &q->filter_list;
416 }
417
418 static struct Qdisc_class_ops prio_class_ops = {
419         .graft          =       prio_graft,
420         .leaf           =       prio_leaf,
421         .get            =       prio_get,
422         .put            =       prio_put,
423         .change         =       prio_change,
424         .delete         =       prio_delete,
425         .walk           =       prio_walk,
426         .tcf_chain      =       prio_find_tcf,
427         .bind_tcf       =       prio_bind,
428         .unbind_tcf     =       prio_put,
429         .dump           =       prio_dump_class,
430         .dump_stats     =       prio_dump_class_stats,
431 };
432
433 static struct Qdisc_ops prio_qdisc_ops = {
434         .next           =       NULL,
435         .cl_ops         =       &prio_class_ops,
436         .id             =       "prio",
437         .priv_size      =       sizeof(struct prio_sched_data),
438         .enqueue        =       prio_enqueue,
439         .dequeue        =       prio_dequeue,
440         .requeue        =       prio_requeue,
441         .drop           =       prio_drop,
442         .init           =       prio_init,
443         .reset          =       prio_reset,
444         .destroy        =       prio_destroy,
445         .change         =       prio_tune,
446         .dump           =       prio_dump,
447         .owner          =       THIS_MODULE,
448 };
449
450 static int __init prio_module_init(void)
451 {
452         return register_qdisc(&prio_qdisc_ops);
453 }
454
455 static void __exit prio_module_exit(void)
456 {
457         unregister_qdisc(&prio_qdisc_ops);
458 }
459
460 module_init(prio_module_init)
461 module_exit(prio_module_exit)
462
463 MODULE_LICENSE("GPL");